ARTIFICIAL NEURAL-NETWORK VERSUS MULTIPLE LINEAR-REGRESSION - PREDICTING P B RATIOS FROM EMPIRICAL-DATA/

Traditionally, multiple linear regression models (MLR) are used to pre dict the somatic production/biomass (P/B) ratio of animal populations from empirical data of population parameters and environmental variabl es. Based on data from 899 benthic invertebrate populations, we compar ed the prediction of P/B by MLR models and by Artificial Neural Networ ks (ANN). The latter showed a slightly (about 6%) but significantly be tter performance, The accuracy of both approaches was low at the popul ation level, but both MLR and ANN may be used to estimate production a nd productivity of larger population assemblages such as communities.